Excessive d_r_mutex_lock_app spinlock overhead on AArch64

On tracing a proprietary app, DR seems to spend excessive time in the spinlock in d_r_mutex_lock_app.

Changing the value of the flag spinlock_count_on_SMP to zero helps.

This issue is to figure out the reason behind the excessive spinlock overhead on AArch64, as this works well on x86.